There have been many incidents where I have been involved in discussions around Answering Machine Detection(AMD). Lets first try to understand what is AMD and why it deserves a article in this blog. Most of the international Outbound Call Centers (Dialling to US UK etc) face a issue that around 40% – 60% of the numbers usually go to Answering Machines. This can waste a lot of time for the agents hence Predictive dialer usually do something called “Answering Machine Detection”. Objective is simple to filter out the answering machins and give only Human contacts to the agents. How is it done?
Well the dialer actually listens to the call when connected and the AMD algorithm identifies whether its a human being or an answering machine. If its a answering machine it drops the call (or leaves a message) otherwise it transfer the call to the agent.
All this happens in a few seconds (and if it does not the customer hangs up the call and you get a CALL DROP).
Each vendor talks about a detection rate and interestingly there is no definition of answering machine detection %.
Lets assume that during a shift Dialer detected 20 Answering machines and 20 Answering machines went to agents. Most people would jump to the conclusion that Answering machine detection is 50%. I am afraid that is not correct. Apart from the above two numbers it is important to see how many CONNECTED calls where there in total.
Lets take two scenarios.
Scenario A:
(CC) Total Connected Calls (At dialer level) 100
(DD) Calls disposed as AMD by dialer 20
(AD)Calls disposed by Agents as AMD 20
Scenario B:
(CC) Total Connected Calls (At dialer level) 50
(DD) Calls disposed as Answering Machines by dialer 20
(AD)Calls disposed by Agents as Answering Machines 20
By our previous calculation the detection would be the same but actually in Scenario A dialer has done detection on 100 calls and in scenario B it was on 50 calls. Remember all calls that were actually humans and went to agents are also correct detection.
Hence the % mistake that dialer has done is = AD/CC
Scenario A 20%
Scenario B 40%
So Correct Answering Machine detection for both scenarios is:
Scenario A 80% detection
Scenario B 60% detection
One interesting fact is that this does not consider all humans that were detected as answering machines by the dialer (and there is no way to establish that apart from testing it manually for 100s of known numbers)
So next time when a vendor talks about the detction rate do calculate it this way to ascertain.
The crude and correct way to calculate the Answering machine Detection algorithm is
(1-(AD/CC))%
I would like to hear from vendors on this observation and see if we can arrive at a better formula. Any Call Centers who want to share there numbers are also welcome.
